What this threat is
Satellites are infrastructure in the same sense that power grids and internet cables are infrastructure: they're invisible to most people until they stop working, at which point the consequences are immediate and broad. GPS timing signals coordinate financial transactions and power grid synchronization, not just navigation. Military communications, weather forecasting, earth observation, and broadband internet in remote areas all depend on functional satellites. The number of satellites in orbit has grown dramatically in recent years as commercial operators have deployed large constellations, and the dependency of modern life on space infrastructure has grown with them.
AI is enabling a new category of satellite capability called proximity operations, which means spacecraft that can approach, inspect, maneuver around, or interact with other satellites autonomously. The legitimate applications of this capability are real: repairing satellites that would otherwise become debris, refueling them to extend their operational life, and deorbiting defunct hardware to reduce the collision risk. But the same capability that enables repair and servicing enables interference, jamming, spoofing, physical manipulation, and destruction. A spacecraft that can dock with a satellite to refuel it can also interfere with its antenna or physically disable it. The dual-use problem here is not theoretical; it's built into the physics of the capability.
Anti-satellite weapons, or ASATs, have existed since the Cold War, but AI makes them more capable and harder to attribute. Kinetic ASATs, which destroy a satellite by collision, create debris clouds that threaten every satellite in nearby orbits, including the attacker's own assets. Non-kinetic approaches, including jamming communication links, spoofing GPS signals, dazzling optical sensors with high-powered lasers, and cyber attacks on ground control systems, are harder to attribute and don't create the immediate debris problem. AI makes these non-kinetic approaches faster to execute, harder to detect, and more precisely targeted. An AI-enabled jamming or spoofing system can adapt its signals in real time to counter electronic countermeasures, making the attack more persistent and harder to defeat.
The Kessler syndrome is a specific debris-cascade risk that deserves explanation. The basic idea is that if enough objects in a particular orbital band collide, each collision produces debris that increases the probability of further collisions, which produce more debris, in a self-sustaining cascade that could eventually make that orbital band unusable for generations. Several tests of kinetic ASAT weapons have already produced significant debris clouds. Large-scale conflict that involved attacks on satellites in heavily populated orbital bands could create a cascade that affects the entire commercial and military satellite ecosystem for decades. AI-enabled autonomous systems that can conduct rapid, precise attacks on multiple targets are a factor that increases the risk of exactly this kind of scenario.
Why it matters
The dependence of modern civilization on space infrastructure is not well understood by most people because it's mostly invisible. GPS is the clearest example. Most people think of GPS as a navigation aid, but it's also a global timing infrastructure that financial networks use to timestamp transactions, that power grids use to synchronize across long distances, and that telecommunications networks use to coordinate frequency use. A sustained loss of GPS capability wouldn't just mean that people couldn't find directions; it would disrupt financial systems, power distribution, and communication networks in ways that have cascading effects across the economy. Similar dependencies exist for satellite communications and earth observation data.
The impossibility of rapid human decision-making is a fundamental problem for space conflict that AI makes worse. Communication delays between ground control and satellites in low Earth orbit are measured in fractions of a second, which sounds short but is long enough that an AI-enabled system acting autonomously can complete an engagement before a human operator is even aware it's happening. For satellites in higher orbits, the delays are longer, which increases the pressure to delegate decision-making authority to autonomous systems. And unlike a military engagement in a contested airspace over land, where a miscalculation can be quickly communicated and deescalated, a space incident happens in an environment with no direct communication channel between the parties, no witnesses, and no established protocols for what constitutes an act of hostility versus a malfunction or accident.
The escalation risk is particularly acute because of the entanglement between commercial and military space assets. Commercial satellite constellations are operated by private companies but serve military customers, host military payloads, and provide services that militaries have come to depend on. Targeting those satellites could be framed as targeting civilian infrastructure, military support infrastructure, or both simultaneously, creating ambiguity about the appropriate response that is dangerous in a crisis. The proliferation of satellite constellations run by actors from multiple countries, including actors with genuinely different legal frameworks and operational norms for space, means that misidentification, misattribution, and misinterpretation of intent are all realistic risks. AI systems making rapid autonomous decisions in this environment have limited ability to resolve that ambiguity before acting.
Where things stand today
Military space programs in multiple countries are actively developing AI-enabled autonomous satellite capabilities, proximity operations vehicles, and enhanced ASAT systems. Commercial satellite proliferation has accelerated dramatically, with several major constellations now operating thousands of satellites in low Earth orbit, and more planned. The same trend that's making satellite broadband broadly available is also populating orbital bands with potential targets and potential collision risks at a pace that existing space traffic management infrastructure wasn't built to handle.
The Outer Space Treaty of 1967 is the foundational governance framework for space, and its limitations in the current environment are significant. It prohibits placing nuclear weapons in space and establishing military bases on the Moon or other celestial bodies, but it doesn't prohibit conventional weapons in space, doesn't address autonomous AI systems operating in orbit, and doesn't establish any mechanism for resolving disputes about whether a particular satellite action constitutes an attack. The treaty was negotiated in a world where space activities were limited to a small number of state actors with very few assets; it wasn't designed for a world with thousands of commercial satellites and AI-enabled proximity operations vehicles operated by both states and private companies.
New governance efforts are underway but far from adequate. Several countries have adopted voluntary norms around destructive ASAT testing, committing not to test kinetic weapons that create long-lived debris clouds. Space traffic management is being discussed in multilateral forums, with proposals for registries, communication protocols, and coordination mechanisms. How AI specifically should be governed in space contexts, including what decisions should be reserved for human operators versus what can be delegated to autonomous systems, is receiving some attention in defense and academic circles but hasn't been addressed in any binding international agreement. The gap between the pace of capability development and the pace of governance development is as wide in space as it is in any other AI-related domain.
How Better Societies helps
Summit: Space governance and AI governance are two fields that don't naturally communicate with each other, but the risks at their intersection require people from both communities to be in the same conversation. The Better Societies Summit creates cross-sector dialogue on space situational awareness, AI autonomy in space systems, and the governance frameworks that responsible space powers need to build before incidents force the issue. We bring together space policy experts, AI governance researchers, defense analysts, and commercial operators who all have a stake in getting this right.
Compliance: AI systems used in aerospace and space applications face specific obligations under the EU AI Act's high-risk classification framework. Safety-critical aerospace AI requires conformity assessments, documentation of the AI system's intended use and limitations, human oversight provisions, and robustness testing. Our compliance programs help aerospace AI developers and operators understand their obligations under the Act and build the evidence-based compliance documentation that regulators will expect.
Accelerator: Space situational awareness tools that give operators better visibility into what's happening in orbit, collision avoidance systems that can operate reliably without creating autonomous engagement risks, and verification tools that can help distinguish accidental proximity events from deliberate interference are all areas where technical innovation could significantly reduce risk. If you're building in the intersection of AI and space security, the Better Societies Accelerator connects you with the policy community, the defense research community, and the commercial space operators who need the kind of trusted, verifiable tools you're building.